I'm using BioPython 1.53 with Python 2.6. The following code was working until the recent EFetch updates:
handle = Entrez.efetch(db="nucleotide", rettype="gb", id=seq)
where 'seq' is simply a string with an accession number. Now, however, nothing is being returned. I'm not sure what I need to fix as none of the changes seem to affect what I've coded above.
•
written
by
Sabrewolfy •
In addition to the multiple ID change behavior, NCBI also changed some database names and the default return modes. Here is a working query with Biopython 1.58:
from Bio import Entrez
import urllib2
Entrez.email="test@test.com"
handle = Entrez.efetch(db="nuccore", rettype="gb", retmode="text", id=76096369)
print handle.readline()
try:
handle = Entrez.efetch(db="nuccore", rettype="gb", retmode="text",
id='wrong')
except urllib2.HTTPError:
print "Bad id"
except IOError:
print "Problem connecting to NCBI"
Should give:
LOCUS NM_007726 5807 bp mRNA linear ROD 19-FEB-2012
Bad id
Thanks. I have tried changing the db and rettype details, but it still does not work with BioPython 1.53. I will try with your specific example.
Tested again with "nuccore" but it does not work with BioPython 1.53. Some change NCBI has made has broken this in 1.53 completely. None of the work-arounds have solved it.
1.53 is quite old now so there may have been bug fixes over the past several releases. Could you upgrade to 1.58 and retry? If that doesn't work, knowing the ID that is failing for you could help us reproduce the problem.
@Brad Chapman: Thanks, I have manually upgraded to 1.59 and adjusted my code as per your example above. I notice the handle object no longer has peekline which I was using, so I'll have to fix that, but I think the fetching problem is solved now.
I was using the length of peekline to determine if a sequence had in fact been returned or not (to check if an invalid accession number had been provided, for example). I'm fetching only one sequence.
Although I see now that Entrez.efetch raises an HTTPError if an invalid accession number if given.
But there is no HTTPError to catch in a try clause.
Yes, Peter has put in a lot of work to make the error handling more transparent so you don't have to manually check for problems. I added example code for catching and identifying bad records and network errors. Hope this helps.
Thanks for the example. I tried the urllib2.HTTPError, but forgot to import urllib2 :)
Log in to answer this question.
Can you please open a bug report on biopython's bug tracker? https://redmine.open-bio.org/projects/biopython/issues?set_filter=1&tracker_id=1
Thanks for the report. This is due to some changes at NCBI and is fixed in the current codebase: http://lists.open-bio.org/pipermail/biopython/2012-February/007743.html There will be a new release in the next week or so with these changes included.
@Brad Chapman: Thanks for the link. Didn't come acros that during my searching.
I'm still not clear on why this is not working for me. In the example, 'seq' contains just a string for one sequence -- I am not passing in a list for multiple sequences. However, it still does not work. I've tried the various solutions mentioned in the link above (nuccore/fasta and the
joinsolution to create a comma-separated list), but neither work. My understanding is that the URL change should have no effect if only ONE sequence is requested.The NCBI changes report that "EFetch URLs with multiple IDs must be entered as: id=1,2,3" and "EFetch no longer accepts invalid URL parameters, e.g., id=1&id=2&id=3". However, if only one sequence is requested, the URL would be the same ... it would end with
id=1.As per Brad's full answer below, the problem is the NCBI changed the default retmode.